Always explicitly load the kpse module

[Witiko]

Closes #3. Compared to the workaround given in #3, this solution is automatic for the user and does not require setting a kpse global.

[zauguin]

First of all we can't call kpse.set_program_name unconditionally without breaking kpse for other usecases since it resets the program name.

Beside that it feels like working around a conscious decision of ConTeXt. IMO doing something like local kpse = kpse or require'kpse' would make sense since this could deal with the case that kpse isn't loaded, but here the existing kpse library is explicitly hidden and overwritten with a stub. That ConTeXt decided not just not to use kpathsea but to actively hide it indicates to me that it shouldn't be used in that context. Does anyone know if the library still being available though require is an intended feature of ConTeXt or if it was just forgotten to disable it there?

[zauguin]

This breaks name lookup when used in LaTeX.

[Witiko]

How so?

[zauguin]

It changes the current progname in kpathsea to kpsewhich from this point on, therefore all engine specific overwrites in texmf.cnf are disabled and expansions of $progname in lookup paths lead to the wrong result.
So if you for example have a tex file with the same name in texmf/tex/luatex and texmf/tex/generic, then the luatex version is preferred before and the generic one afterwards.

This doesn't matter too much with TeX Live's default setup since that always falls back to the other directories if a file isn't found and TeX Live generally doesn't reuse filenames, but it can easily break user settings.

[Witiko]

Here is my understanding: ConTeXt is commonly distributed as a part of the ConTeXt Standalone distribution, which does not contain either kpathsea or other parts of TeX Live. If you need kpsewhich in TeX Live, you can explicitly require it but it is unnecessary for the working of ConTeXt.

[Witiko]

However, I tend to agree that the responsibility for loading kpathsea in ConTeXt should perhaps fall on the shoulders of users, seeing how it affects the global state. Therefore, you may want to close both this pull request and #3 as a wontfix.

[Witiko]

@zauguin Out of curiosity, where is tex.initialize defined? I don't see any mention in either luatex.pdf or ltluatex.pdf.